Augmented Collaborative Filtering Recommendation with Implicit Negative Feedback

Abstract

Thanks to the advance in technology there are many kinds of online services for us Among them e-commerce audio and video streaming are the most popular ones These platforms provide a huge amount of goods for customers but customers also feel confused to choose something from these choices Thus we need a recommendation system to help us to find out what we might need or like Traditional recommendation systems utilize user’s explicit information like the member profile and the score rated by users It is hard to collect great amount of this kind of data because people are lazy to give feedback Moreover the performance of recommendation will decrease seriously due to the sparsity of dataset To solve this problem more and more researches start to infer users’ preference by analyze their behavior Many companies start to collect user’s behavior records such as clicking and browsing time We get this kind of data by observing user’s actions that are easier to collect than explicit feedback It is called implicit feedback because we don’t know the preference extent of users from this data In this paper we propose a ranking based recommendation system We rank the items by comparing the items with users’ feedback and we suppose the relation between positive and implicit negative feedback is more feasible than the relation between the same kind of feedbacks Thus we utilize this relation as the basis of our ranking method to improve the accuracy of our recommendation Eventually we increase the performance of recommendation and obtain the result which is more discriminative than the other approaches with implicit feedback

Date of Award	2017 Sept 12
Original language	English
Supervisor	Hung-Yu Kao (Supervisor)